Overview
Aura is a local-first, provider-neutral AI agent platform designed for ongoing work on infrastructure controlled by its operator. A single Go binary provides the agent runtime, tools, command-line interface, Telegram gateway, and embedded web cockpit, while each user receives a separate identity, memory database, workspace, and sandbox.
The surrounding Docker Compose stack supplies services for storage, graph memory, embeddings, document ingestion, web search, voice, and bundled integrations. Model inference can use OpenRouter, a ChatGPT plan, the bundled llama.cpp server, or Ollama; cloud providers receive submitted context, while local inference keeps that processing on the host.
Key Features
- Runs a streaming agent loop with shared step and time budgets plus repeated-call controls.
- Discovers deferred tools and loads schemas when needed, including tools exposed by mounted MCP servers.
- Stores facts, sources, and validity windows in a temporal graph, with supporting evidence returned for memory paths.
- Retrieves indexed document passages with source hashes and citations while retaining access to original files.
- Provides terminal and filesystem tools with approval gates for destructive commands and secret redaction.
- Supports skill creation, bundled memory and messaging integrations, scheduled jobs, and additional MCP servers.
Architecture and Data
Aura keeps conversations authoritative in Postgres and builds a derived recall projection with managed context compaction. ArcadeDB supplies graph-native memory, allowing remembered facts to include provenance and time validity rather than existing only as unstructured chat history.
The platform separates storage location from inference location. Data can remain on self-hosted services, but privacy during model execution depends on whether the selected model provider is local or cloud-based.
Typical Uses
Aura fits multi-user, self-hosted agent deployments that need persistent memory, document-grounded answers, recurring work, or messaging access. Its tool system also supports agents that must work with files and terminals, extend themselves through skills, or connect to external capabilities through MCP.